From 125c19a3c31a6290d7feea3faf41dd20fb97e2f4 Mon Sep 17 00:00:00 2001
From: hedaoyuan <hedaoyuan@baidu.com>
Date: Tue, 8 Nov 2016 15:35:57 +0800
Subject: [PATCH] fix some nvcc compile options (#392)

---
 CMakeLists.txt                | 15 +++++++++++++++
 cmake/util.cmake              |  8 --------
 paddle/gserver/CMakeLists.txt |  2 +-
 paddle/math/CMakeLists.txt    |  2 +-
 4 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 282e3e199..39f876bc9 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -95,11 +95,26 @@ if(NOT WITH_GPU)
     add_definitions(-DHPPL_STUB_FUNC)
     list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
 else()
+    if(${CUDA_VERSION_MAJOR} GREATER 6)
+        if(COMPILER_SUPPORT_CXX11)
+            LIST(APPEND CUDA_NVCC_FLAGS -std=c++11)
+        endif()
+    endif()
+
     # TODO(yuyang18): Change it to remove std=c++11 in cuda compile.
     set(CUDA_PROPAGATE_HOST_FLAGS OFF)
     if(NOT CUDNN_FOUND)
         message(FATAL_ERROR "Paddle need cudnn to compile")
     endif()
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-g -O3 --use_fast_math")
+
+    if(WITH_AVX)
+        if(AVX_FOUND)
+            set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler -mavx")
+        endif(AVX_FOUND)
+    else(WITH_AVX)
+        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler -msse3")
+    endif(WITH_AVX)
 
     if(WITH_DSO)
         set(CUDA_LIBRARIES "")
diff --git a/cmake/util.cmake b/cmake/util.cmake
index 3f78cd08c..a8282f071 100644
--- a/cmake/util.cmake
+++ b/cmake/util.cmake
@@ -188,14 +188,6 @@ macro(add_simple_unittest TARGET_NAME)
     add_unittest(${TARGET_NAME} ${TARGET_NAME}.cpp)
 endmacro()
 
-macro(add_paddle_culib TARGET_NAME)
-    set(NVCC_FLAG ${CUDA_NVCC_FLAGS})
-    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--use_fast_math)
-    cuda_add_library(${TARGET_NAME} STATIC ${ARGN})
-    set(CUDA_NVCC_FLAGS ${NVCC_FLAG})
-endmacro()
-
-
 # Creates C resources file from files in given resource file
 function(create_resources res_file output)
     # Create empty output file
diff --git a/paddle/gserver/CMakeLists.txt b/paddle/gserver/CMakeLists.txt
index 9ac4d210f..a066f80c2 100644
--- a/paddle/gserver/CMakeLists.txt
+++ b/paddle/gserver/CMakeLists.txt
@@ -50,7 +50,7 @@ if(NOT WITH_PYTHON)
 endif()
 
 if(WITH_GPU)
-    add_paddle_culib(paddle_gserver ${GSERVER_SOURCES})
+    cuda_add_library(paddle_gserver ${GSERVER_SOURCES})
 else()
     add_library(paddle_gserver STATIC
         ${GSERVER_SOURCES})
diff --git a/paddle/math/CMakeLists.txt b/paddle/math/CMakeLists.txt
index db305812a..93b1bf46a 100644
--- a/paddle/math/CMakeLists.txt
+++ b/paddle/math/CMakeLists.txt
@@ -23,7 +23,7 @@ if(NOT WITH_GPU)
     add_library(paddle_math STATIC
         ${MATH_SOURCES})
 else()
-    add_paddle_culib(paddle_math ${MATH_SOURCES})
+    cuda_add_library(paddle_math ${MATH_SOURCES})
 endif()
 
 
-- 
GitLab